is-this-ai-generated

[is-this-ai-generated]

and did you have to look it up

[is-this-ai-generated]

PSA about pings to posts

Hi there. I wanted to clarify something about how I handle pings to posts, presumably with the intent to have me identify the post image as genAI or not.

generally speaking, i would prefer if you at least first make an attempt to ask the original poster about the source of the image before sending me anything. if you don't receive an answer within a reasonable amount of time, then you can send me an ask with a link to the post in question rather than ping me in the post comments itself.

why? well, a few reasons.

one, the witch-hunting aspect. for better or worse, people have gotten so cagey about images posted online potentially being genAI that they've started to sling accusations at each other first and ask questions later. for many independent artists–whether digital, traditional, or graphic designers–being accused of using genAI is akin to a scarlet letter. pinging a blog like mine to a post's comments is basically telling the poster that you don't believe that they're being genuine. if you have a suspicion that the image is genAI, just ask the OP first.

two, pinging me–or other accounts like mine–directly to a post's comments can inspire the OP to lock down their account or delete their posts altogether. some might say that's a "victory" against genAI users, but i don't. why? because of the price of potential collateral damage: you could inadvertently run a genuine artist off the platform, simply because you couldn't be arsed to ask them upfront.

three, my blog is an educational/informative blog, not a call-out blog. i take as neutral a standpoint as i possibly can when i post. and i did not create this blog to shame anyone for any reason.

in fact, the impetus behind me creating this blog was seeing tons upon tons of botted accounts abusing tags to label their very-clearly genAI images as "photography." and they would get thousands of notes of people asking things like "where is this? i wanna go!" or "i wish i knew where this photo was taken."

it was at that moment i realized just how difficult it was for the average person to spot genAI and that immediately became a concern of mine. not because of tag abuse, not even because of passing off a genAI image as "art." no, it worries me because of the risk of people falling for scams, deepfakes, propaganda, etc.

so i wanted to make this blog to help prime those who might be interested in learning how to figure it out on their own. the intent was not to make people overly-suspicious or even paranoid, but rather to gently remind one to exercise healthy skepticism and cultivate an inquisitive mindset. it's why i post my process as to how and why i came to the conclusion that i reached. it's also why i don't go into an OP's comments and shout from the rooftops "PSA: THIS IS GENAI!" i simply reblog it from the last user i come across that has also reblogged it and add my notes there.

bottom line: getting in the habit of questioning everything you see posted online without sources, especially things you enjoy looking at, find pretty, or interesting, will help better prepare you to think critically about other, more important things. that's the ethos of this blog.

so here's my easy guide to how to handle finding an image post you're suspicious of being genAI:

check the tags on the post.

check the source of the image. if one isn't provided, do a tineye or reverse google images search.

if you still can't find anything, respectfully ask the OP of the post if it's genAI. just be up-front and don't be a dick, even if you loathe genAI.

if they don't answer for days or block you outright, you can send me a link via my ask box and i'll take a look.

as always, feel free to send me any questions you may have about genAI, either on tumblr or off tumblr. to be absolutely clear, i don't want to discourage people from asking questions. i just operate this blog with a certain level of decorum that i would like to upkeep.

thanks!

[is-this-ai-generated]

AI generated verdict: provable and confirmed

long post incoming. sorry (not actually sorry).

tells: this was a tough one, at first glance, to point out any one aspect that made me go "that's definitely genAI." chiefly, the subject matter was that special kind of uncanny that triggered my brain to look deeper.

i couldn't find the smoking gun on a cursory investigation with just my eyes. nothing like "that guy's got 7 fingers and an extra arm" here. as such, i speculated that it was done by a very proficient user of genAI, someone who has a lot of experience in crafting prompts and iterating on output.

but that didn't stop me because the uncanny feeling was pretty intense.

in the end, the major things that tipped me off were meta-contextual in nature, if we were to assume this was a real photograph: the practicality of the object, and, crucially, the presumed level of media coverage of the object, were it to exist in real life.

much like with the egg salad machine post from last year, it boils down to two questions: "would this object feasibly exist in the real world? and if so, would it cause media stir/considerable attention if it did?"

put it simply: if you see an image purporting itself to be a real photo and your immediate reaction is "wow! that's really cool, i've never seen anything like that before in my life! it's really impressive, even!" my tip is to reverse-image-search it, to be safe.*

(*if you care, that is. all my posts are written with the assumption that you're reading them because you want to avoid getting tricked by genAI.)

and if, by image-searching said image, you are unable to find any article, no matter how big or small, highlighting its existence, and yet it's somehow all over social media, you can allow yourself to feel suspicious. because it is.

so in my search, predictably, i could not find any such article anywhere about this supposed object. but i did find the inspiration that this image seems to be based around.

there appeared to be a set of arcade cabinets that were carved from marble mentioned in a handful of niche gaming publications and/or other dubiously verifiable outlets in 2023... except it wasn't actually carved from marble. it was an earlier instance of genAI images for proof of concept purposes.

on to the source of the image itself.

source: it required a bit of deep-diving in multiple reverse-image searching platforms to drill down to the original poster.

eventually, i was able to uncover the following account on twitter: https://twitter.com/goo_vision

further than that, i was able to find the original tweet in which they posted this image: https://twitter.com/goo_vision/status/1909308836143362139

this account is self-identified as an "artist and designer" and states that they use "generative AI tools" in their bio. the poster also appears to get a fair amount of AI haters in their comments, to which they've replied defending their use of genAI in their works. lastly, they also post multiple images that all share a very similar theme repeatedly within a short period of time (such as 20+ images of various types of "arcade or media cabinets made of unusual materials"), a hallmark of iterative genAI images.

why is that a hallmark? when you request a genAI create an image based on a text string/prompt, it will often produce multiple feasible images that it "thinks" match to what you asked for. the user can then rate each image, if they so choose, to help educate/train the algorithm to be more accurate in the future. it looks as though this user may be using midjourney (or something similar; i'm guessing midjourney because it's really good at replicating photorealism) to generate these images and are picking through the ones they like best. pretty standard stuff, nothing out of the ordinary there.

bottom line is that this is another example of someone on tumblr reposting something they found elsewhere and not only not giving appropriate backlinks to where they found it, but also failing to tag it appropriately as genAI.

alright, this post is long enough. if you have questions about this one, or any other posts in general, feel free to send me a message; asks are open.

and as always, my ever-present disclaimer for my blog: I do not make these posts to drum up drama, call for brigade-ing/trolling of any identified sources, or insinuate any sort of moral failing on anyone whatsoever for reblogging/liking these images. i also do not make these posts to "dunk on" genAI users or enjoyers.

I created this blog to educate those who want to learn about what to look out for when it comes to AI generated images, how I come to my conclusions, and the tells they can look out for in the future. I do not condone any measure of harassment, bullying, or hate mail sent to the originating artist(s) that may be identified in my posts. this should also not be used as a jumping off point to harass or send hate mail to any blog I reblog from, either.

please check my pinned post for more info on how I run this blog.

[is-this-ai-generated]

AI generated verdict: moderately likely

upon searching the credited artist, their website pops up. on their "about" page, this paragraph can be found:

For many years, I was a perfectionist, pouring [sic] over each detail for weeks on end on a single piece. In early 2024 I shifted my approach into a daily creation practice, embracing a more fluid process. Now, I let each concept unfold with minimal overthinking, sourcing material from paintings and illustrations in the Public Domain, photography, 3D renders, AI, and my own sketches.

(emphasis added by me)

however, it's difficult to say the percentage of AI usage in this piece or to what degree it was used. in looking at the detail of the picture, there are a few artifacts that seem to point to at least partial generation, such as with the fingers.

the ring finger of the left hand shows a telltale sign of AI generation in that it seems as though the AI algorithm got "confused" about where to put the knuckle or tip of the finger. part of the ring finger indicates it should be bent, thus the very last part of the digit would be obscured. however, as many algorithms do, the subtle details possibly became lost in the overall process. this is common; AI algorithms are very good at basic shapes, especially when there isn't a lot of detail to be generated. however, as more detail is requested, more mistakes crop up. the AI has to quickly calculate a best guess as to where to put what for it to match the overall prompt and level of detail requested.

all of this to say that the artist themselves has admitted to using AI as a tool for their collages on their portfolio site, which is more than I can say of a lot of the blogs on tumblr that tag their genAI pictures as "photography." I'm definitely not annoyed.

once again, I must stress: I do not make these posts to drum up drama or insinuate any sort of moral failing on anyone whatsoever. I created this blog to educate those who want to learn about what to look out for when it comes to AI generated images, how I come to my conclusions, and the tells they can look out for in the future. I do not condone any measure of harassment, bullying, or hate mail sent to either the originating artist or any blog here that I may reblog from. this should also not be used as a jumping off point to harass or send hate mail to any blog I reblog from, either. please check my pinned post for more info on how I run this blog.

Chema Mendez - mendezmendez

[is-this-ai-generated]

AI generated verdict: provable and confirmed

tells: credited account, when searched, points to an instagram account wherein the artist self-identifies as an "AI explorer" and has many similar pieces that are labeled as generated art. all art on the account has similar types of stylistic choices, though it is clearly iterated on several times to minimize as many obvious tells as possible. since instagram imposes a rule that all AI generated content must be labeled, they have done so, to their credit.

this is a perfect example of how someone who is skilled and practiced using a specific algorithm can iterate on a piece numerous times until virtually all obvious tells are eliminated. further, the style that was chosen mimics that of "corporate advertising art" of the 80s and early 90s, of which there are many examples of flat colors and hard, sharp edges, or very simple, centralized subjects (i.e.: the images are not very "busy" and the backgrounds are often soft gradients or buildings). AI algorithms can nail down these particular styles with exceptional accuracy, as it requires less "guesswork" on their part.

however, there are some minor tells still visible. the curvature of (presumably) the earth has an odd raised portion that does not coincide to any logical purpose (i.e.: it appears to be a sun rising, but there seems to be a portion where the algorithm got "confused" and duplicated the arc of light to the right of the sun). also on the right-hand side of the image, we can see a clear delineation between the majority of the image and a small, triangular portion of the planet. this does not seem to serve any function or purpose; again, it appears as though the algorithm got "confused" and placed a hard edge there.

as this piece is no longer showing on the credited user's instagram account, it is possible that they noticed this disparity and removed it. it's not clear.

it is important to reiterate that this blog is not meant to be used as a jumping off point to harass any OPs whose images are shown here. i don't want anyone to be using this as an excuse to troll. this blog is purely for the purpose of education so that those who are wary of AI imagery are more equipped to identify it and make an informed decision on the type of content they would like to support.

“Awareness is wakefulness. Wakefulness is consciousness. Consciousness is limitless awareness.” – Frederick Lenz

Limitless Blank Embrace @BlankEmbrace

[is-this-ai-generated]

AI generated verdict: highly likely

tells: badge on machine appearing to state "EGSG SALAD SAMDWICH", letters on badge seems blurry or incomplete, man's left and right hand/fingers not proportioned correctly, positioned correctly, and/or anatomically incorrect or impossible, man's left leg positioned in an impossible perspective (i.e.: should be nudging directly against the machine but appears to be phasing through it), button on the farthest left appears to be unfinished/not fully rendered, wood paneling slats appear unevenly sized in a way that does not happen in real wood paneling, various artifacts in image that seem to indicate unfinished rendering, subject matter is uncanny, lighting direction does not match between objects.

[is-this-ai-generated]

Lynx fascinated by a hummingbird...

[is-this-ai-generated]

welcome to "Is This AI Generated?"

hi there! i have been seeing a lot of reblogs lately of AI generated art being passed around as real photography or real art. this blog serves as a place of learning, where you can ask questions or submit art you feel may be AI generated.

some key things to keep in mind:

don't use my blog as a jumping-off point for a harassment campaign. i don't fuck widdat.

you can submit any type of art that you may think is AI generated, but please ensure it complies with tumblr's ToS. i ain't here to get my blog flagged as mature for no raisin.

when submitting, please make sure you provide your reasoning as to why you think the image might be AI generated, if the image was credited or not (and if that credit goes to a real person), and where you saw it. reference links are highly appreciated.

please keep questions and asks in general on-topic.

if you're not sure if i've answered a question before, check the tags for "answered." i'd prefer if this was done prior to sending the ask.

if I have reblogged a post from your blog and you're coming here to see what this is all about, please understand that I do not reblog anything with the intent to be making a "statement" or "judgement" about your character or moral standing. that's not what this blog is for, nor is it what it's about in the slightest. I don't know your life and if you're into AI art, more power to you. this is purely a neutral place of learning.

In that same vein, I'm not here to make anyone feel stupid or dumb for not recognizing a piece or image as AI art. it's oftentimes insanely difficult to tell when the original image source has used an image generator, especially if the prompter is well-versed on iterative prompts with AI algorithms. someone with advanced knowledge of how to use these generators can easily make a very convincing image after performing some iterations; not every prompter will be satisfied with the very first image an AI produces. and for that matter, not every image of an anime character will have seven fingers or flippers for hands or other obvious tells. ergo, if you see something here that you couldn't tell as being AI, don't feel dumb!

and if you reblogged anything seen here, don't feel bad! I'm here to help you learn, if you so desire.

policy on pinging me to post comments directly.

FAQ.

if i run into any other key points that need highlighting, i will update this post! it should always be pinned, so be sure to check it before submitting or asking.

thanks!

[is-this-ai-generated]

why "poisoning" AI algorithms simply doesn't work how you think it does.

off and on, i'll come across posts that get tens of thousands of reblogs that are peppered with alarmist and urgent language, saying things like "if you're an artist, do this to your art now" and so on. they tend to link to things like nightshade or glaze.

while these tools are great for perhaps long-term defense and offense against AI scraping, they will do nothing to protect art that has already been scraped. this is something i see overlooked entirely when people make a mad rush to utilize this tool.

when an algorithm is provided a data set, that data set is typically a snapshot at the moment it was scraped. this means that if there are changes made to that data set online, they more than likely will not affect the current data set in use in real time. algorithms work in sandbox environments to prevent confusing results for this reason.

this means that if your art has been scraped and you go back and nightshade and/or glaze the pieces, then replace the old file with the new file, it doesn't matter. that data has already been processed.

however, that's not to say it's still not a decent idea anyhow for subsequent scrapes made by other algorithms. it's not a bad idea to experiment with them. but that of course depends on your computer's ability to even use these tools and if your art style doesn't suffer from the results of the "poisoning."

the system requirements to run nightshade and glaze are as follows:

your PC must be running Windows 10 or 11 OR MacOS. Linux is not supported.

to get the most out of using nightshade or glaze, you must have a dedicated NVIDIA GPU on a win10 or win11 machine. this means that if you are using a computer that only has integrated graphics (i.e.: no dedicated GPU), OR your system uses an AMD GPU, your experience will be vastly different.

as mentioned, that dedicated GPU must be an NVIDIA GPU to use GPU rendering mode. they recommend one of the GPUs on this list. you will want to look under the RTX and/or GEforce banners to locate your card. the rest are workstation units.

there are known bugs with specific NVIDIA GPUs, specifically in the GTX 1660/1650/1550 lines. this may affect your results.

if you do not have a dedicated GPU, or are using an AMD GPU, you can still use nightshade or glaze, but it will devour your PC's CPU resources. this means you will experience extreme system slow-down at best, or a system crash at worst if your CPU isn't fast enough or have enough cores. as such, you have to make sure that there are no other programs running when using these tools.

MacOS installs appear to be limited to ARM CPUs, meaning M1, M2, and M3 chipsets. if your mac uses an intel processor, you're out of luck.

some other things to consider when using nightshade or glaze:

they seem to cite 20 minutes as an average time elapsed for "poisoning" a piece of artwork. this might be affected by the file size and the speed of the CPU or GPU being used.

since this is essentially rendering software, your PC may run hot if it's under-powered. make sure your PC has adequate ventilation.

if you have a lot of art to "poison," this might take you hours if not days to fully complete.

mobile devices are not supported (i.e.: no app you can download) and do not appear to be on the roadmap at all at this time.

images of certain textures may end up distorted to the human eye. this seems to be primarily for flat colors or smooth backgrounds. this means if your art is generally flat-shaded, you will likely see distortion on the final image.

glaze can be used on mobile devices, but only on the web browser. again, a mobile app does not appear to be on the roadmap at this time. this method is also invite-only, though they seem generous with their invites.

the functionality of using these tools is very involved and might be overwhelming to those who have no experience or comfort with technology. this isn't as simple as "upload and it does the rest for you," at least in the case of nightshade and the desktop version of glaze.

nightshade and glaze are algorithms in and of themselves. yes, they themselves are essentially AI. this means that they need to be constantly updated to match the advancements of the algorithms they're combating against.

because of this, it's highly likely that when updates are made to the algorithms for glaze and nightshade, you will need to re-glaze or "re-poison" your art as those updates are made.

since both tools are made by the university of chicago, they rely on grants for funding. as we know, grant providers can be fickle and funding could be removed at any time. i wasn't able to find any sort of donation funding option, but that doesn't mean they don't accept donations. however, it would still be dependent on the department of the university that houses this project.

i still see people passing around posts about how glazing and/or nightshade-ing your art is a surefire way to "defeat the AI bros," but it just isn't that simple, unfortunately. it also isn't especially "user friendly" if you aren't already comfortable with navigating programs on a desktop or laptop environment. further than that, it also seems to benefit a specific subset of computer users, primarily those who use windows machines and NVIDIA CPUs. basically, if you keep up-to-date with PC gaming as well as do your own art, you'll probably be able to use these tools no problem.

but that's not to say this isn't a step in the right direction. if you are able to use these tools, i would highly recommend them, especially if you are a very prolific professional artist, someone who gets a lot of commissions, or someone who is selling your art on merchandise in some capacity.

and if you have discovered an algorithm that's using your art style? get a lawyer; you may have a case against the algorithm in use under DMCA. but y'know. i'm not a lawyer so. ask one. yeah.